Forget call centers, local energy prices mean Britain's latest offshoring wave is AI projects
Brit firms look to run tech overseas as govt tries to support 'sovereign' creators
One in five UK firms have already moved AI workloads abroad due to high energy costs, in findings likely to alarm a government counting on AI to drive economic growth.
The warning is contained in the "Land, Power, Compute" report published by rent-a-GPU firm CUDO Compute, which naturally has skin in this game.
The UK is a standout case, but the data - compiled by Censuswide, which canvassed senior AI decision makers in 700 organizations - spans the US and Europe too. The study reveals how the cost and availability of power is reshaping where AI infrastructure gets built, and where AI applications will ultimately live.
Billions spent racing to acquire GPUs and cloud capacity are running into a hard constraint: in many places, the facilities to host them can't be built fast enough, held back by planning delays and slow grid connections.
In Santa Clara, California - home of Nvidia - nearly 100 MW of newly constructed datacenter capacity is sitting empty, it claims, awaiting power connections that may not come for years.
Energy costs are a particular concern in Britain, which has some of the highest prices in the developed world. And for AI infrastructure specifically, the impact is stark: a third of UK organizations say energy costs are limiting their ability to scale.
Among AI-first companies, nearly a third (32 percent) indicate they are considering moving workloads overseas due to power costs - a direct threat to the UK government's ambitions to make Britain a global AI leader and use the technology to drive economic recovery.
Yet around 30 percent are prioritizing sovereign or regionally controlled compute, even if it comes at a higher cost, showing that organizations do want to build in the UK but need the infrastructure to be there, CUDO claims.
To try to counter the UK's crippling energy costs, the government says it is taking action to break the influence of gas on electricity prices. Because gas-fired power stations are called upon last to balance the grid, such as when wind or solar output is low, they determine the final price of electricity.
This week, the government detailed new measures including long term fixed contracts, and proposed raising the Electricity Generator Levy – or windfall tax - to ensure a larger proportion of the extraordinary revenues generated when the gas price spikes is available to government to support businesses and households.
These measures and others are intended to incentivize generators to move on to fixed contracts not linked to volatile gas, the government claims.
The US is rated as the most attractive location for new AI cluster provisioning, with three-quarters of respondents viewing it positively, followed by India. Eastern Europe also scores highly, with 58 percent seeing it as a positive location for AI deployment, above Western Europe on 45 percent, and the Nordics (44 percent). China also ranks highly with 55 percent of respondents thinking it a good location for AI projects.
Xiantao datacenter in Chongqing, China – Pic credit: QinJin/ Shutterstock
"AI sovereignty is being hotly discussed as a priority for UK organizations, but it only works if the infrastructure exists to support it," claimed CUDO Compute CEO Matt Hawkins. "What we are seeing is a growing tension between where businesses want to run AI and where they actually can."
The problem isn't unique to the UK. Across the US and Europe, the same bottleneck recurs: land, power, cooling, compute, and permits rarely align in the same place at the same time. Even where governments are trying to accelerate delivery - the UK has streamlined planning and other incentives - physical infrastructure keeps falling behind demand.
- You too can build a nuclear battery from junk you have lying around the house
- Trump-branded datacenter project fails to make itself great, again
- AI is reshaping Britain's datacenter map away from London
- UK.gov kicks off half-a-billion quid sovereign AI venture with £80M invite
Infrastructure - power, networking and other energy related inputs, rather than GPU pricing - now drives most of the cost in AI deployment. This is reshaping where AI gets built: site selection is increasingly less about proximity to end users and more about proximity to available power, as The Register reported this week. Regions with spare grid capacity and available land are winning investment over established datacenter hubs.
Cudo says nearly one in five survey respondents have already scaled back AI workloads due to energy costs, and more than a fifth claim energy bills account for more than a third of their AI infrastructure budgets.
A third of respondents are actively evaluating alternative regions for AI deployment due to infrastructure considerations, and roughly a quarter claimed they would relocate entirely to secure renewable power.
Organizations need to get all their ducks in a row when planning AI deployments and the infrastructure needed to support them, CUDO says - treating zoning, power procurement, and cooling engineering as core competencies rather than downstream logistics handled as an afterthought. ®
